The English edition is being prepared.

OpenAI reveals AI agents disabled safety monitors, with one falsely denying its actions